NP-COMPLETE PROBLEMS
Admin Two more assignments… No office hours on tomorrow
Run-time analysis We’ve spent a lot of time in this class putting algorithms into specific run-time categories: O(log n) O(n) O(n log n) O(n 2 ) O(n log log n) O(n 1.67 ) … When I say an algorithm is O(f(n)), what does that mean?
Tractable vs. intractable problems What is a “tractable” problem?
Tractable vs. intractable problems Tractable problems can be solved in O(f(n)) where f(n) is a polynomial
Tractable vs. intractable problems What about… O(n log log log log n )? O(n 100 )?
Tractable vs. intractable problems Technically O(n 100 ) is tractable by our definition Why don’t we worry about problems like this?
Tractable vs. intractable problems Technically O(n 100 ) is tractable by our definition Few practical problems result in solutions like this Once a polynomial time algorithm exists, more efficient algorithms are usually found Polynomial algorithms are amenable to parallel computation
Solvable vs. unsolvable problems What is a “solvable” problem?
Solvable vs. unsolvable problems A problem is solvable if given enough (i.e. finite) time you could solve it
Sorting Given n integers, sort them from smallest to largest. Tractable/intractable? Solvable/unsolvable?
Sorting Given n integers, sort them from smallest to largest. Solvable and tractable: Mergesort: Θ (n log n )
Enumerating all subsets Given a set of n items, enumerate all possible subsets. Tractable/intractable? Solvable/unsolvable?
Enumerating all subsets Given a set of n items, enumerate all possible subsets. Solvable, but intractable: Θ (2 n ) subsets For large n this will take a very, very long time
Halting problem Given an arbitrary algorithm/program and a particular input, will the program terminate? Tractable/intractable? Solvable/unsolvable?
Halting problem Given an arbitrary algorithm/program and a particular input, will the program terminate? Unsolvable
Integer solution? Given a polynomial equation, are there integer values of the variables such that the equation is true? Tractable/intractable? Solvable/unsolvable?
Integer solution? Given a polynomial equation, are there integer values of the variables such that the equation is true? Unsolvable
Hamiltonian cycle Given an undirected graph G=(V, E), a hamiltonian cycle is a cycle that visits every vertex V exactly once A B E D F
Hamiltonian cycle Given an undirected graph G=(V, E), a hamiltonian cycle is a cycle that visits every vertex V exactly once A B E D F
Hamiltonian cycle Given an undirected graph G=(V, E), a hamiltonian cycle is a cycle that visits every vertex V exactly once A B E D F
Hamiltonian cycle Given an undirected graph G=(V, E), a hamiltonian cycle is a cycle that visits every vertex V exactly once A B E D F
Hamiltonian cycle Given an undirected graph, does it contain a hamiltonian cycle? Tractable/intractable? Solvable/unsolvable?
Hamiltonian cycle Given an undirected graph, does it contain a hamiltonian cycle? Solvable: Enumerate all possible paths (i.e. include an edge or don’t) check if it’s a hamiltonian cycle How would we do this check exactly, specifically given a graph and a path?
Checking hamiltonian cycles
Make sure the path starts and ends at the same vertex and is the right length Can’t revisit a vertex Edge has to be in the graph Check if we visited all the vertices
NP problems NP is the set of problems that can be verified in polynomial time A problem can be verified in polynomial time if you can check that a given solution is correct in polynomial time (NP is an abbreviation for non-deterministic polynomial time)
Checking hamiltonian cycles Running time? O(V) adjacency matrix O(V+E) adjacency list What does that say about the hamilonian cycle problem? It belongs to NP
NP problems Why might we care about NP problems? If we can’t verify the solution in polynomial time then an algorithm cannot exist that determines the solution in this time (why not?) All algorithms with polynomial time solutions are in NP The NP problems that are currently not solvable in polynomial time could in theory be solved in polynomial time
P and NP P NP Big-O allowed us to group algorithms by run-time Today, we’re talking about sets of problems grouped by how easy they are to solve
Reduction function Given two problems P 1 and P 2 a reduction function, f(x), is a function that transforms a problem instance x of type P 1 to a problem instance of type P 2 such that: a solution to x exists for P 1 iff a solution for f(x) exists for P 2 f x f(x) P 1 instance P 2 instance
Reduction function Where have we seen reductions before? Bipartite matching reduced to flow problem All pairs shortest path through a particular vertex reduced to single source shortest path Why are they useful? f x f(x) P 1 instance P 2 instance
Reduction function f Problem P 2 x f(x) yes no yes no Problem P 1 Allow us to solve P 1 problems if we have a solver for P 2 f x f(x) P 1 instance P 2 instance answer
Reduction function f Problem P 2 x f(x) P 2 solution Problem P 1 f ’ P 1 solution Most of the time we’ll worry about yes no question, however, if we have more complicated answers we often just have to do a little work to the solution to the problem of P 2 to get the answer
Reduction function: Example P1 = Bipartite matching P2 = Network flow f Problem P 2 x f(x) P 2 solution Problem P 1 f ’ P 1 solution Reduction function (f): Given any bipartite matching problem turn it into a network flow problem What is f and what is f’?
Reduction function: Example P1 = Bipartite matching P2 = Network flow f Problem P 2 x f(x) P 2 solution Problem P 1 f ’ P 1 solution Reduction function (f): Given any bipartite matching problem turn it into a network flow problem A reduction function reduces problems instances
NP-Complete A problem is NP-complete if: 1. it can be verified in polynomial time (i.e. in NP) 2. any NP-complete problem can be reduced to the problem in polynomial time (is NP-hard) The hamiltonian cycle problem is NP-complete What are the implications of this? What does this say about how hard the hamiltonian cycle problem is compared to other NP-complete problems?
NP-Complete A problem is NP-complete if: 1. it can be verified in polynomial time (i.e. in NP) 2. any NP-complete problem can be reduced to the problem in polynomial time (is NP-hard) The hamiltonian cycle problem is NP-complete It’s at least as hard as any of the other NP-complete problems
NP-Complete A problem is NP-complete if: 1. it can be verified in polynomial time (i.e. in NP) 2. any NP-complete problem can be reduced to the problem in polynomial time (is NP-hard) If I found a polynomial-time solution to the hamiltonian cycle problem, what would this mean for the other NP-complete problems?
NP-complete If a polynomial-time solution to the hamiltonian cycle problem is found, we would have a polynomial time solution to any NP- complete problem Take the input of the problem Convert it to the hamiltonian cycle problem (by definition, we know we can do this in polynomial time) Solve it If yes output yes, if no, output no f Ham-Problem: P 2 x f(x) yes no yes no NP problem NP problem answer
NP-complete Similarly, if we found a polynomial time solution to any NP-complete problem we’d have a solution to all NP- complete problems f Solved NP-Problem: P 2 x f(x) yes no yes no NP problem NP problem answer
NP-complete problems Longest path Given a graph G with nonnegative edge weights does a simple path exist from s to t with weight at least g? Integer linear programming Linear programming with the constraint that the values must be integers
NP-complete problems 3D matching Bipartite matching: given two sets of things and pair constraints, find a matching between the sets 3D matching: given three sets of things and triplet constraints, find a matching between the sets Figure from Dasgupta et. al 2008
P vs. NP Polynomial time solutions exist NP-complete (and no polynomial time solution currently exists) Shortest path Bipartite matching Linear programming Minimum cut … Longest path 3D matching Integer linear programming Balanced cut …
Proving NP-completeness A problem is NP-complete if: 1. it can be verified in polynomial time (i.e. in NP) 2. any NP-complete problem can be reduced to the problem in polynomial time (is NP-hard) Ideas?
Proving NP-completeness Given a problem NEW to show it is NP-Complete 1. Show that NEW is in NP a. Provide a verifier b. Show that the verifier runs in polynomial time 2. Show that all NP-complete problems are reducible to NEW in polynomial time a. Describe a reduction function f from a known NP-Complete problem to NEW b. Show that f runs in polynomial time c. Show that a solution exists to the NP-Complete problem IFF a solution exists to the NEW problem generate by f
Proving NP-completeness Show that a solution exists to the NP-Complete problem IFF a solution exists to the NEW problem generate by f Assume we have an NP-Complete problem instance that has a solution, show that the NEW problem instance generated by f has a solution Assume we have a problem instance of NEW generated by f that has a solution, show that we can derive a solution to the NP-Complete problem instance Other ways of proving the IFF, but this is often the easiest
Proving NP-completeness Why is it sufficient to show that one NP-complete problem reduces to the NEW problem? Show that all NP-complete problems are reducible to NEW in polynomial time
Proving NP-completeness All others can be reduced to NEW by first reducing to the one problem, then reducing to NEW. Two polynomial time reductions is still polynomial time! Show that all NP-complete problems are reducible to NEW in polynomial time
Proving NP-completeness Show that all NP-complete problems are reducible to NEW in polynomial time Show that any NP-complete problem is reducible to NEW in polynomial time Show that NEW is reducible to any NP-complete problem in polynomial time BE CAREFUL!
NP-complete: 3-SAT A boolean formula is in n-conjunctive normal form (n-CNF) if: it is expressed as an AND of clauses where each clause is an OR of no more than n variables 3-SAT: Given a 3-CNF boolean formula, is it satisfiable? 3-SAT is an NP-complete problem
NP-complete: SAT Given a boolean formula of n boolean variables joined by m connectives (AND, OR or NOT) is there a setting of the variables such that the boolean formula evaluate to true? Is SAT an NP-complete problem?
NP-complete: SAT 1. Show that SAT is in NP a. Provide a verifier b. Show that the verifier runs in polynomial time 2. Show that all NP-complete problems are reducible to SAT in polynomial time a. Describe a reduction function f from a known NP-Complete problem to SAT b. Show that f runs in polynomial time c. Show that a solution exists to the NP-Complete problem IFF a solution exists to the SAT problem generate by f Given a boolean formula of n boolean variables joined by m connectives (AND, OR or NOT) is there a setting of the variables such that the boolean formula evaluate to true?
NP-Complete: SAT 1. Show that SAT is in NP a. Provide a verifier b. Show that the verifier runs in polynomial time Verifier: A solution consists of an assignment of the variables If clause is a single variable: return the value of the variable otherwise for each clause: call the verifier recursively compute a running solution polynomial run-time?
NP-Complete: SAT Verifier: A solution consists of an assignment of the variables If clause is a single variable: return the value of the variable otherwise for each clause: call the verifier recursively compute a running solution linear time -at most a linear number of recursive calls (each call makes the problem smaller and no overlap) -overall polynomial time
NP-Complete: SAT Show that all NP-complete problems are reducible to SAT in polynomial time a. Describe a reduction function f from a known NP-Complete problem to SAT b. Show that f runs in polynomial time c. Show that a solution exists to the NP-Complete problem IFF a solution exists to the SAT problem generate by f Reduce 3-SAT to SAT: - Given an instance of 3-SAT, turn it into an instance of SAT Reduction function: DONE -Runs in constant time! (or linear if you have to copy the problem)
NP-Complete: SAT -Assume we have a 3-SAT problem with a solution: -Because 3-SAT problems are a subset of SAT problems, then the SAT problem will also have a solution -Assume we have a problem instance generated by our reduction with a solution: -Our reduction function simply does a copy, so it is already a 3-SAT problem -Therefore the variable assignment found by our SAT-solver will also be a solution to the original 3-SAT problem Show that a solution exists to the NP-Complete problem IFF a solution exists to the NEW problem generate by f Assume we have an NP-Complete problem instance that has a solution, show that the NEW problem instance generated by f has a solution Assume we have a problem instance of NEW generated by f that has a solution, show that we can derive a solution to the NP-Complete problem instance
NP-Complete problems Why do we care about showing that a problem is NP- Complete? We know that the problem is hard (and we probably won’t find a polynomial time exact solver) We may need to compromise: reformulate the problem settle for an approximate solution Down the road, if a solution is found for an NP-complete problem, then we’d have one too…
CLIQUE A clique in an undirected graph G = (V, E) is a subset V’ ⊆ V of vertices that are fully connected, i.e. every vertex in V’ is connected to every other vertex in V’ CLIQUE problem: Does G contain a clique of size k? Is there a clique of size 4 in this graph?
CLIQUE A clique in an undirected graph G = (V, E) is a subset V’ ⊆ V of vertices that are fully connected, i.e. every vertex in V’ is connected to every other vertex in V’ CLIQUE problem: Does G contain a clique of size k? CLIQUE is an NP-Complete problem
HALF-CLIQUE Given a graph G, does the graph contain a clique containing exactly half the vertices? Is HALF-CLIQUE an NP-complete problem?
Is Half-Clique NP-Complete? 1. Show that NEW is in NP a. Provide a verifier b. Show that the verifier runs in polynomial time 2. Show that all NP-complete problems are reducible to NEW in polynomial time a. Describe a reduction function f from a known NP-Complete problem to NEW b. Show that f runs in polynomial time c. Show that a solution exists to the NP-Complete problem IFF a solution exists to the NEW problem generate by f Given a graph G, does the graph contain a clique containing exactly half the vertices?